Research On Image Fusion Based On Regional Characteristics

Image fusion is a technology that extracts available information from two images or more to merge into an image which contains richer information about the scene and is more consistent with the visual effects of human.Image fusion technique has been widely used in remote sensing,medical diagnostics,computer vision and other fields.At present,image fusion technology is mainly based on pixel-level fusion and regional-level fusion.The former does not analysis the regional characteristics of the source images,and uses consistent fusion rules which makes the contrast and resolution of final fusion image low;the latter has improved the quality of the fusion image,but most of the algorithm divide the scene into significant and non-significant regions,though the regional-level fusion method makes significant area of the fusion image more prominent,the non-significant area is blurred.This paper systematically studies the basic theory and frontier algorithms of image fusion,focusing on infrared and visible image fusion and multi-focus image fusion.The main work and contributions of the paper are as follows:1 The basic theory of image fusion and the related frontier algorithms of infrared and visible image fusion and multi-focus image fusion are studied.2.An algorithm of multi-focus image fusion method based on defocus depth estimation and NSCT is proposed.In this algorithm,an approach of region division based on defocus estimation is presented.Firstly,the defocus depths of source images are estimated;secondly,the initial focus detection map is obtained by comparing the defocus depth estimation images of the source images;thirdly the initial focus detection map is divided into the focus region and the indefinite region by mathematical morphology filtering.Then different fusion rules are adopted to fuse disparate regions according to diverse regional characteristics.Experiments show that this method can extract the focus area more accurately,and has better versatility and stability than the image segmentation algorithm based on the traditional visual attention model.The experimental comparison validates that the fusion algorithm has a pretty good performance both on subjective and objective quality evaluation.3.A fusion algorithm of infrared and visible image based on regional characteristics is proposed.Firstly,multi-target characteristics of infrared image have been considered while region growth and guided filter are employed to extract targets adaptively,thus the locations of targets are obtained by simple regional growth.Secondly the filter radius of the guided filter has been improved;the local regions of target are obtained by filtering with adaptive window,whose size is decided by the size of the target.At the same time the background information is effectively eliminated.Then the final target map is obtained by regional growing.Thirdly the gradient map of the visible image is calculated,and the visible image is divided into smooth region and texture region by the k-means clustering and the gradient map.After that,the final scene division map composed of the target region,the smooth region and the texture region is obtained by combining the target extraction map and division map of visible image.Finally,this paper designs different fusion rules according to different regional characteristics.The combination of regional characteristics of the image and pixel-level fusion method can fully present the information of the fused image,which solves the related problems such as the poor visual effect of the fusion image and detailed information is not rich enough in traditional fusion method based on the NSCT.The experimental results show that the proposed method has a good advantage in that it can not only enhance the contrast and clarity of the fusion image but also highlight the target characteristics of the fusion image.